Is there any known electromagnetic / interference / [no idea] effect which can cause IEEE 802.11 (Wi-Fi) 2.4GHz devices located in very close proximity (one on top of the other) to severly interfere with each other, even though they are using completely non-overlapping channels of the available ISM 2.4GHz spectrum?
The general knowledge about 2.4GHz Wi-Fi suggests that networks operating on non-overlapping channels should not interfere: answer at Network Engineering explaining 2.4GHz Wi-Fi channels and their interference
2.4GHz Wi-Fi channel structure: 
Having a network topology like this (dots mean wireless, dashes are ethernet wire):
laptop . . . router A ----- router B . . . router C ----- ISP's box (FTTH ONT)
(client (access (client) (access
to A) point) to C point)
Router A's network uses channel 6, router C's network was tested on both channels 1 and 13 (13 gives even more clearance than needed and is legal to use in my country).
If routers A and B are placed physically on each other, the maximum achievable throughput to/from the ISP's speedtest server is about 4Mbps, which would normally resemble an overcrowded channel or channel with many collisions. However, everyone seems to claim that networks in my setup should not interfere as they are assigned non-overlapping channels with enough clearance.
If I move router A a few meters away from B (not sure how much minimum distance would be enough though), the throughput rockets to about 20Mbps, which is around the expected maximum of both wireless networks and means they both perform close to their maximum capacity.
No other changes in either configuration or topology are made on any of the devices, the only change that affects the throughput is physically moving the router A away from B.
Tested multiple times over multiple days to rule out short-term conditions (neighbors), in two different rooms to rule out environment/furniture, no other wireless clients connected in either network during testing, no microwaves in the vicinity, no issues with bandwidth to/from ISP (100/100 Mbps fiber connection achieving full speed to the ISP's speedtest server when on ethernet wire).
Any ideas?
Other notes:
- This happens on two different vendors' hardware which should fully conform to Wi-Fi specifications, so it should not be an issue with a single vendor or hardware piece or any non-standard 802.11 protocol modifications.
- I'm not using the wider 802.11n 40MHz channels, this happens on pure 802.11g-only (22MHz) settings too.
Best Answer
It depends on the internal architecture of the receivers in each box - these are designed to match a specification and down to a price.
A radio receiver has two functions - selectivity, and amplification - and a high performance receiver will achieve each function in several stages - alternating between selectivity and gain.
For example, a tuned filter, followed by an RF amplifier, followed by a tuned filter, followed by a mixer (which often has gain), followed by an IF filter, followed by an IF amplifier, and so on.
Note that the main selectivity (channel rejection) is provided by the IF filter, and the front end RF filters will pass the entire 2.4 GHz band, but protect the following amplifier from other strong signals.
A lower cost receiver may omit the first tuned filter ahead of the RF amplifier. It will actually be more sensitive because that filter had some attenuation. However, extremely strong signals - even outside the 2.4GHz band altogether - can overload the receiver, distorting the wanted signals and mixing in components of the unwanted signals.
That is essentially what is happening here : because the boxes are stacked on top of each other, one's transmission is seen by the next as an extremely strong interfering signal, (in the 2.4GHz band, though outside the desired channel) overloading the receiver.
You may find that Brand B's receiver is more tolerant to overload than Band A's, but really, without getting deeply into the receiver design there's not a lot you can do about in except prevent the overload, by physically separating the boxes.